How to minimize artifacts in atomistic simulations of membrane proteins, whose crystal structure is heavily engineered: beta-2-adrenergic receptor in the spotlight

How To Minimize Artifacts in Atomistic Simulations of MembraneProteins, Whose Crystal Structure Is Heavily Engineered:β2‑Adrenergic Receptor in the SpotlightMoutusi Manna,† Waldemar Kulig,† Matti Javanainen,† Joona Tynkkynen,† Ulf Hensen,‡ Daniel J. Muller,‡

Tomasz Rog,† and Ilpo Vattulainen*,†,§

†Department of Physics, Tampere University of Technology, P.O. Box 692, FI-33101 Tampere, Finland‡Department of Biosystems Science and Engineering (D-BSSE), ETH-Zurich, 4058 Basel, Switzerland§MEMPHYS-Center for Biomembrane Physics, University of Southern Denmark, Odense, Denmark

*S Supporting Information

ABSTRACT: Atomistic molecular dynamics (MD) simulationsare used extensively to elucidate membrane protein properties.These simulations are based on three-dimensional proteinstructures that in turn are often based on crystallography. Theprotein structures resolved in crystallographic studies typically donot correspond to pristine proteins, however. Instead thecrystallized proteins are commonly engineered, includingstructural modifications (mutations, replacement of proteinsequences by antibodies, bound ligands, etc.) whose impact onprotein structure and dynamics is largely unknown. Here weexplore this issue through atomistic MD simulations (∼5 μs intotal), focusing on the β2-adrenergic receptor (β2AR) that is oneof the most studied members of the G-protein coupled receptor superfamily. Starting from an inactive-state crystal structure ofβ2AR, we remove the many modifications in β2AR systematically one at a time, in six consecutive steps. After each step, weequilibrate the system and simulate it quite extensively. The results of this step-by-step approach highlight that the structuralmodifications used in crystallization can affect ligand and G-protein binding sites, packing at the transmembrane-helix interfaceregion, and the dynamics of connecting loops in β2AR. When the results of the systematic step-by-step approach are compared toan all-at-once technique where all modifications done on β2AR are removed instantaneously at the same time, it turns out thatthe step-by-step method provides results that are superior in terms of maintaining protein structural stability. The results providecompelling evidence that for membrane proteins whose 3D structure is based on structural engineering, the preparation ofprotein structure for atomistic MD simulations is a delicate and sensitive process. The results show that most valid results arefound when the structural modifications are reverted slowly, one at a time.

■ INTRODUCTION

G-protein coupled receptors (GPCRs) constitute one of thelargest families of plasma membrane receptors.1,2 They aremajor contributors to the cellular signal transduction andrespond to a wide variety of extracellular stimuli such as light,taste, odor, peptides, neurotransmitters, and hormones. Duringextracellular signaling, the ligand-induced change in receptorconformation propagates through the receptor’s transmem-brane region to its cytoplasmic domain. An activated GPCRthen couples with its intracellular G-protein (guaninenucleotide-binding protein) to transmit a biochemical signalto the cytosolic side. Importantly, as GPCR signaling regulatesmajor physiological processes, GPCRs are associated with anumber of diseases and have therefore become the key targetfor drug discovery.1

Although the growing numbers of available crystal structureshave significantly advanced our understanding of GPCRs, thesestatic structures are insufficient to describe the dynamic features

that govern GPCR function. Moreover, a matter of concern isthe fact that the crystal structures are mostly resolved in non-native detergent media3 and are often highly engineered toovercome the inherent conformational flexibility of receptors.4,5

The concept of engineering here refers to structuralmodifications often used to foster GPCR crystallization.4,5

Commonly used techniques include, e.g., introduction ofstabilizing mutations to the protein structure, ways to bindantibody fragments, or fusion of T4-lysozyme to flexibledynamic loop regions. Given that such engineering is acommon strategy in membrane protein structure determina-tion, there is a growing need to understand the impact of suchstructural modifications on the relationship between receptorstructure and function.6

Received: January 26, 2015

Article

pubs.acs.org/JCTC

© XXXX American Chemical Society A DOI: 10.1021/acs.jctc.5b00070J. Chem. Theory Comput. XXXX, XXX, XXX−XXX

pubs.acs.org/JCTC

http://dx.doi.org/10.1021/acs.jctc.5b00070

http://pubs.acs.org/action/showImage?doi=10.1021/acs.jctc.5b00070&iName=master.img-000.jpg&w=228&h=117

Currently, the understanding of potential issues arising fromstructural modifications in protein structure determination isvery limited. This largely stems from inherent difficulties tocharacterize receptor behavior over nanoscales, where theresolution of most experimental techniques is no longersufficient to gauge atom-scale changes in protein structureand dynamics. The method of choice to unravel this issue isatomistic molecular dynamics (MD) simulations. Atom-scalesimulations have become highly powerful techniques tocomplement experiments in studies of protein complexes.Recent progress in the field has rendered studies of highlycomplex proteins possible over time scales where even slowchanges in conformational states can be observed.7 Togetherwith homology modeling, atomistic MD simulations cancontribute to structure prediction, and more generally theycan provide substantial insight into rational drug design andreceptors’ conformational dynamics.7−12 As more crystalstructures have become available, computational efforts toexplore GPCRs and clarify their properties in atomistic detailhave become common.Here we use atomistic MD simulations to assess the impact

of structural modifications used in protein structure determi-nation on the structure, structural stability, and dynamics of theβ2-adrenergic receptor (β2AR), which is one of the moststudied members of the GPCR superfamily. β2AR is expressedin pulmonary and cardiac myocyte tissues and is a therapeutictarget for asthma and heart failure.2 GPCRs, including β2AR,have attracted considerable attention since a great fraction ofdrug development uses GPCRs as the main target.To elucidate the effects of structural modifications, we study

β2AR through two complementary approaches. In the firstapproach (Method A, “A” standing for “all”), starting from theengineered crystal structure of β2AR, we remove all of thenumerous structural modifications at the same time, therebyreverting the receptor back to its pristine structure, and thenexplore the protein behavior in long MD simulations. In thesecond approach (Method S, “S” standing for “stepwise”), againstarting from the engineered crystal structure of β2AR, thestructural modifications used in experimental structuredetermination are removed in six successive steps, one at atime. After each of the steps in Method S, β2AR is simulatedquite extensively to clarify the impact of every structuralmodification separately. In the end of the procedure, we againend up with the pristine structure where all modifications areremoved, and the final model of β2AR is simulated to compareits behavior with that of Method A and with the structurepredicted by experimental structure determination.13

Our results show that, although the stepwise approach(Method S) is time-consuming and computationally moreexpensive than the more straightforward approach of revertingall structural modifications at once (Method A), Method Sproduces superior results in terms of structural protein stability.The take-home message of the present work is that thepreparation of membrane protein structure for atomistic MDsimulations is a very delicate and sensitive process. Great care isrequired in order to avoid possibly significant artifacts that mayarise if the engineered protein structure is reverted for atomisticsimulations without a sufficiently rigorous process.The results of the step-by-step modifications in Method S

also provide relevant insight into the effects of crystallizationmodifications on the intrinsic dynamics of the apo-receptor:this approach shows how structural modifications used forcrystallization can affect ligand and G-protein binding sites,

packing at the transmembrane helix interface, and the dynamicsof connecting loops.

■ METHODS AND MODELSExperimentally Determined Structure of β2AR. Due to

recent breakthroughs in development of crystallizationtechniques, structures of the inverse agonist-bound β2ARwere resolved recently.13−15 These structures exhibit thecommon architecture of GPCRs: seven membrane-spanninghelices connected to each other by intracellular andextracellular loops (see the cartoon diagram of β2AR in FigureS1 (Supporting Information (SI)). Although β2AR and otherclass-A GPCRs such as rhodopsin share structural homology,they have noticeable differences, too. For instance, as comparedto inactive dark-state rhodopsin, inactive β2AR exhibits arelatively open structure due to a weaker interaction betweenthe cytoplasmic ends of the transmembrane (TM) helixes 3 and6 and a broken salt-bridge in the highly conserved (E/D)RYmotif.14 These differences may account for the higher basalactivity and structural instability of β2AR. In 2011, agonist-bound crystal structures of β2AR were published: here, theactive conformation of the receptor was stabilized either by theintercellular partner G-protein or a G-protein mimetic nanobody.16,17 The subtle agonist-induced rearrangement in theligand-binding pocket of β2AR move TM5 around Ser207inward by about 2 Å. This conformational change at theextracellular side leads to a much larger conformational changeat the distal G-protein binding site, including an outward 14 Åshift of the cytoplasmic end of TM6. These differences providethe structural basis of the allosteric mechanism of the receptoractivation.4

In this study, we concentrate on the inactive form of β2AR[PDB id: 3D4S], whose crystal structure does not correspondto the pristine receptor.13 Rather, the crystal structure is for areceptor that is quite heavily engineered. First, the receptor iscocrystallized with the partially inverse agonist timolol. Second,it has several structural modifications such as a missingintracellular loop 3, a fused T4-lysozyme between TM5 andTM6, and multiple mutations.13 Here using atomistic MDsimulations for this receptor structure, we explore what are themost efficient and valid ways of removing the structuralmodifications used in crystallization.In the first approach (Method A), we simultaneously remove

all the changes made for crystallization and then allow thereceptor (in its apo-form) to evolve in a lipid bilayerenvironment. In the second approach (Method S), we firstequilibrate the crystal structure in the membrane environmentand then remove the many structural modifications one-by-onein a number of successive steps, each followed by hundreds ofnanoseconds for equilibration. The findings of the present workare compared with a wealth of available experimental data.

Lipid Membrane Structure. A lipid bilayer in a fluid(liquid-disordered) state, used in the present studies, wasinitially comprised of 288 DOPC (1,2-dioleoyl-sn-glycero-3-phosphocholine) molecules. The lipid membrane structure wasequilibrated for 100 ns before embedding the β2AR; for detailswith regard to the force field and the simulations, see theDiscussion below. After incorporation of the receptor (seebelow) and removal of overlapping lipids, the system contained202 DOPC lipid molecules.To further validate the β2AR model, we performed an

additional control simulation where β2AR was placed in abilayer environment containing about 10 mol % cholesterol.

Journal of Chemical Theory and Computation Article

DOI: 10.1021/acs.jctc.5b00070J. Chem. Theory Comput. XXXX, XXX, XXX−XXX

B


This choice was based on the conditions used in ref 13 todetermine the structure of β2AR: the study used a ratio of 10%w/w cholesterol/monoolein lipid in the cubic phase, and forthermal stability assays the study used various differentconcentrations of a cholesterol analogue called cholesterylhemisuccinate (CHS). Meanwhile, it is known that cholesterolmodulates membrane physical properties and is required for theproper functioning of many membrane proteins,18−21 and it hasbeen suggested or shown to influence the functional state andthus the structural properties of β2AR.

13,15,22,23 Therefore, toaccount for these features, we carried out a control simulationwith a membrane comprised of 304 DOPC and 34 cholesterolmolecules that was pre-equilibrated for 100 ns beforeembedding the protein into the lipid bilayer.For incorporating the protein into a mixed lipid bilayer, we

followed our recently published method, where the idea is topush proteins into a lipid membrane from its side by applying ahigh lateral pressure on the system.24

Protein Structure: Description of Structural Modifica-tions Made in Experiments for Crystallization. The initialcoordinates of human β2AR (residues 32−342) that we used inthe simulations were obtained from the PDB id 3D4S.13 Allnonprotein molecules except for the partially inverse agonisttimolol and crystallographic water molecules were disregardedfrom the structure. The X-ray crystal structure13 consideredhere is a complex comprised of β2AR together with a T4-lysozyme (T4L): for structure determination, the residues231−262 of the intracellular loop 3 (ICL3) had beensubstituted by a T4-lysozyme (T4L). The β2AR-T4L complexwhose structure had been resolved consisted of seventransmembrane helices (TMs) (TM1 − residues 32−59;TM2 − residues 67−96; TM3 − residues 103−136; TM4 −residues 147−171; TM5 − residues 197−229; TM6 − residues267−298; and TM7 − residues 305−327), three extracellularloops (ECLs) (ECL1 − residues 97−102; ECL2 − 172−196;and ECL3 − 299−304), three intracellular loops (ICLs) (ICL1− residues 60−66; ICL2 − residues 137−146; and ICL3replaced by T4L discussed above), and a short C-terminal helixH8 (residues 330−340) parallel to a membrane surface (FigureS1). The same definitions of the protein segments will be usedthroughout the manuscript. The receptor’s crystal structure hastwo mutations: (i) E1223.41W on TM3, which in experimentswas introduced to enhance receptor stability, and (ii) a N187Emutation on ECL2 introduced in experiments to avoidglycosylation. The superscripts used in describing β2ARresidues refer to the Ballesteros−Weinstein residue number-ing,25 where the first digit ranging between 1 and 7 refers to thetransmembrane helix number at which that particular aminoacid is located, and the second number after a decimal indicatesits position with regard to the most conserved residue of thathelix, conventionally assigned to a value of 50. This numberingscheme is used here, too. The following section describes themethods we used to remove the structural modifications usedin crystallization in order to model the apo-receptor in its nativestate.Modeling Intracellular Loop 3. For ICL3, the sequence

of β2AR was taken from the UniProt database26 with the accessnumber P07550. We first built an initial conformation of themissing loop attached to the receptor using the MODELER9.11 program27 and then refined the loop using the samesoftware. ICL3 was modeled as an unstructured loopconsidering that its sequence shows evidence for intrinsicdisorder.28 Out of ∼500 models generated, assessment was

made based on DOPE (Discrete Optimized Potential Energy)scoring of MODELER and also based on the loop’s structureand its orientation toward the membrane.

Force Fields and Simulation Details. To parametrize theinteractions in the simulation systems, the all-atom optimizedpotentials for liquid simulation (OPLS) force field29,30 was usedin conjunction with the recently refined parameters for lipidmolecules.31 The TIP3P model, which is compatible with theOPLS parametrization, was used for water molecules.32

Parameters for timolol were built from the standard OPLSforce field, while partial charges were derived using RESP33

fitting to the electrostatic potential according to the OPLSmethodology. All simulations were performed with theGROMACS 4.6.0 software package.34 Simulations were donein the isobaric−isothermal (NpT) ensemble using a time stepof 2 fs and 3D periodic boundary conditions. The v-rescalethermostat35 with a time constant of 0.1 ps was employed tomaintain the temperature at 310 K. The temperatures of thesolute and the solvent were controlled independently. Thepressure for the system (1 bar) was controlled semi-isotropically using the Parrinello−Rahman barostat36 with a 1ps time constant. The LINCS algorithm37 was applied topreserve hydrogen covalent bond lengths. The Lennard−Jonesinteractions were truncated at a cutoff distance of 1.0 nm. Forthe long-range electrostatic interactions, the particle meshEwald method38 was employed with a real space cutoff of 1.0nm, β-spline interpolation (order of 6), and a direct sumtolerance of 10−6. In analysis, error bars were estimated throughstandard error, calculated by dividing the standard deviation ofa given data set with the square root of its sample size.23 Weused the g_analyze tool of GROMACS for error estimation.

Methods Investigated To Remove Structural Mod-ifications Used in Experimental Structure Determina-tion. We used two approaches to remove the modifications.These approaches were then used in separate MD simulations,and their results were compared to one another to find howsignificantly the deformation of the protein structure(compared to experimental structure) depends on the approachused.

Method A: All Changes Made at Once. In this procedure,starting from the engineered crystal structure of the receptor,we applied the commonly used approach, that is, we removedall the structural modifications at the same time, before placingthe apo-receptor in a membrane for equilibration. Therefore,we removed the ligand (timolol), removed the T4-lysozyme,and replaced it with the missing ICL3 and removed mutations.We did not attempt to model the unresolved N-terminal (32residues) and C-terminal (71 residues) parts. After all changes,the receptor was embedded into a pre-equilibrated DOPCbilayer. Subsequently, the system was solvated with 34,516water molecules. 94 sodium and 102 chloride ions were addedto neutralize the net charge of the system and to create a 150mM NaCl solution. After energy minimization, we simulatedthe system for 25 ns with position restraints on protein heavyatoms, letting the other components equilibrate first. This wasfollowed by another equilibration period of 25 ns with positionrestraints focused only on the protein backbone. Aftersimulating for 50 ns, all restraints were released, and theequilibrated system was subjected to a production simulationover a time scale of 1000 ns (referred to as the “Long”simulation). Additionally, in the analysis we also used thetrajectories of three shorter (independent) simulations thatcovered a scale of 200 ns each. The first shorter trajectory



C


(referred to the “Short-1” simulation) covered the first 200 nsof the Long simulation. Then Short-2 and Short-3 were basedon independent simulations of the same system and lasted alsofor 200 ns.Method S: Stepwise Approach. In the second procedure,

the structural modifications used in experimental structuredetermination were removed in a number of successive steps,one at a time. Each step was followed by simulation of thereceptor in a DOPC bilayer. The final structure of the receptorobtained after each step was used as a starting point for theimmediate next step.Below, “EM-PS” refers to energy minimization, followed by

equilibration for 50 ns with restraints acting on the protein(using a procedure as described above) followed by a 200 nsproduction simulation.• Step 1. The β2AR-T4L complex, together with its bound

ligand and mutations, was placed into a DOPC bilayer. 34496water molecules were added to the system. The β2AR-T4Lcomplex prior to any modifications had a charge of +12. 94sodium, and 106 chloride ions were added to neutralize thecharge of the system and to maintain the physiological-like saltconcentration. EM-PS was carried out.• Step 2. The partially inverse agonist timolol was removed

from the extracellular ligand-binding cavity of the β2AR-T4Lcomplex, followed by EM-PS.

• Step 3. T4-lysozyme was removed from the engineeredβ2AR-T4L complex, and the 32 missing residues (residues231−262) of ICL3 were incorporated in between TM5 andTM6. The resulting receptor is referred to as β2AR-ICL3. Thedetails of de novo loop modeling are described above. This steprequired readjustment of the number of neutralizing ions. EM-PS was carried out.• Step 4. W1223.41 on TM3 was mutated back to E1223.41.

We chose the protonated (neutral) state of the E122 residue, asit faced directly toward membrane hydrophobic interior. EM-PS was carried out.• Step 5. E1875.26 on ECL2 was mutated back to N1875.26.

The number of counterions was adjusted (as in Method A to 94Na+ and 102 Cl− ions) and EM-PS was carried out.• Step 6. Finally, the native apo-receptor obtained after the

above-mentioned changes was simulated for 200 ns in a DOPCbilayer and compared with the results of Method A.

Control Simulations of Method S. For completeness, wecarried out two separate tests to confirm that the results forMethod S were on a solid basis:• Control-1. We tested the influence of cholesterol: In

addition to the simulation in a DOPC bilayer, the β2AR modelobtained through Method S (after Step 5) was placed in aDOPC bilayer containing 10 mol % cholesterol, followed byEM-PS. The production simulations lasted for 200 ns.

Figure 1. Properties of the apo-receptor model generated by Method A. A) Root-mean-square deviation (RMSD) of the protein backbone: entireβ2AR (black) and the transmembrane region (cyan) (Long simulation). B) Per residue RMSD (average over the last 100 ns) of the simulatedreceptor model compared to the crystal structure, calculated over the backbone atoms of the transmembrane region (Long simulation). Thesnapshots show the front (TM1-TM4) and back (TM5-TM7) views of the protein’s average structure (the last 100 ns), with TM residues coloredaccording to their respective RSMD values and the rest of the protein represented as a transparent gray cartoon. C) A representative snapshot of thereceptor (from the Long simulation at 900 ns) with TM residues colored according to the secondary structure content (α-helix in magenta, 310-helixin blue, turn in green, coil in orange): the front (TM1-TM4) and back (TM5-TM7) sides. D) The number of residues with helical content (sum ofα-, 310-, and Π-helices) for each TM helix of β2AR obtained from different Method A trajectories (averaged over the last 100 ns): Long (red), Short-1 (blue), Short-2 (green), and Short-3 (magenta). These simulation results were compared to that found in the crystal structure 3D4S (gray). Theerror in the number of residues with helical content was below 0.04.



D



• Control-2. Different sequence of steps: The order of stepsdiscussed above (first removing timolol, then T4L, and finallyreverting the mutations) was chosen based on the expecteddegree of significance: Starting from substantial structuralchanges and moving on structural alterations that wereexpected to be smaller. However, in general the number ofhow these steps can be combined is large, and the results maydepend to some extent on the order in which the structuralmodifications in the protein are reverted. To consider thesensitivity of the order of steps in Method S, we also tested adifferent scenario where the above-mentioned order of steps(Step 2 − Step 5) was reversed. In that case, starting from thefinal snapshot of Step 1, we first removed mutations; second wereplaced the T4-lysozyme with ICL3; and then we removedtimolol. Finally, the native apo-receptor was equilibrated in aDOPC bilayer.As to the final details, the software BODIL 0.8.139 was used

for reverse mutations. The side chain orientation of the back-mutated residues was chosen such that there was no collisionwith its neighboring residues. Prior experimental knowledgewas taken into consideration, where possible. As to theW1223.41 to E1223.41 mutation, the orientation of E1223.41 waschosen to match the one found in another crystal structure ofthe receptor [PDB id: 2RH1].15

■ RESULTS AND DISCUSSION

Below we describe and discuss the structural properties of thereceptor models we have studied. The Discussion focuses onstructural differences arising from different ways to revert theprotein modifications used in experimental crystal structuredetermination.Method A: Reversal of All Crystallization Modifica-

tions at Once Led to Instabilities in Protein Structure.We first used Method A to investigate how the proteinstructure changes during simulations when all modificationsdone for crystallization assays are reverted at once. Here wediscuss the results of the Long simulation, unless statedotherwise. Figure 1A shows the root-mean-square deviation(RMSD) of the protein backbone throughout the simulation.Large RMSD values were observed mainly due to the flexibleintercellular and extracellular loops. RMSD for the trans-membrane part was significantly smaller and reached a plateauwithin ∼90 ns of simulation. The average RMSD between thecrystal structure and the simulation (the last 100 ns), calculatedover the backbone atoms of the transmembrane region, was2.89 ± 0.002 Å.The corresponding RMSD fluctuations (per residue) from

the crystal structure are plotted in Figure 1B. The snapshotshows the average structure (over the last 100 ns) of theprotein with the TM residues colored according to the RMSDvalues. As depicted in Figure 1B, many parts of the proteinunderwent large deviations from the crystal structure. Theseinclude TM1 (especially the N-terminus up to its middle part(RMSD > 5 Å)), TM4, residues 197−207 at the extracellularhalf of TM5 (RMSD > 4 Å), and residues 315−324 in themiddle of TM7 that includes residues from the conservedNPxxY motif. Besides the ends of the TM helices, manyresidues from their center part located at the helical interface orunderneath the ligand-binding cavity deviated considerablyfrom the crystal structure: e.g., residues 115−118 of TM3,residues 282−291 of TM6, residues 315−324 of TM7, etc. Wefound that these deviations were related to a change occurred in

the secondary structure of the protein during the simulation(Figure S2).Figure 1C shows a simulation snapshot of β2AR with the

transmembrane residues colored according to the secondarystructure content. Figure 1D represents the average helicalcontent (over the last 100 ns) of each TM helix calculated withDSSP.40 As shown in Figures 1C and S2, the first 10 N-terminalresidues of the TM1 helix completely unfolded to turn/bendstructure. Partial opening of the helix also occurred at residues42−50 in the middle of TM1. In this part of TM1, we observeda considerable drop in the α-helical content; meanwhile, thepercentage of the 310-helix increased. 310-helix is the proposedintermediate in the folding/unfolding of α-helices and knownto be involved in a helix−coil transition.41 Altogether, theaverage helical percentage of TM1 dropped down to 52%compared to the crystal structure (Figure 1D and Table S1).The unwinding of TM helices also occurred in residues 83−86in the middle of TM2; in residues 162−171 on the extracellularside of TM4 (P168 caused local opening of helical structure);in residues 197−207 in the extracellular-half of TM5 (which isa major part of the ligand-binding pocket); in residues 280−285 in the middle of TM6 and residues 271−274 near theintracellular end of TM6; in residues 315−320 before theNPxxY motif of TM7, etc. (see Figures 1C and S2). As a result,the helical contents decreased to 65% in TM2, 64% in TM4,60% in TM5, 59% in TM6, and 71% in TM7 of their respectivevalues in the crystal structure (Figure 1D and Table S1).In agreement with the results obtained from the Long (1 μs)

simulation, a considerable decrease in helical percentages of theTM domains was also observed in the three short trajectories(Short-1, Short-2, and Short-3 (Figure 1D and Table S1)). Onaverage, in the shorter simulations the most noticeable drop inhelical content occurred for TM1 (58%), TM4 (68%), TM6(56%), and TM7 (62%), as compared to the crystal structure.Summarizing, the β2AR model obtained by Method A, when

simulated in a DOPC lipid bilayer, exhibited considerabledeformation from the crystal structure. The main weakness ofMethod A was observed to be the significant drop in the helicalcontent of TM domains of the receptor. For example, TM1 andTM6 lost almost half of their helicity compared to the crystalstructure. A previous circular dichroism (CD) spectroscopicstudy characterized very highly helical structure (>80%) ofpeptides corresponding to the TM domains of a GPCR (calledthe adenosine A2a receptor) in micelles and lipid vesicles.42

The high helical propensity of the TM domains are required forproper folding of GPCRs43 and thus related to their function.

Method S: Step-by-Step Reversal of All StructuralModifications Was Found to Be a Superior TechniqueCompared to Method A and Resulted in a StableReceptor Model. The structural instability of the receptorobserved in the previous procedure led us to rebuild the modelmore carefully. Here, in Method S we removed the structuralmodifications, done originally for crystallization studies, in astepwise manner. In the following section, we describe theeffects that were observed after reversal of each structuralchange. Finally, when all structural modifications have beenreverted, we compare the properties of the receptor found inthis manner to the structure found through Method A.

Step 1: Equilibration of the Ligand Bound β2AR-T4LComplex in a Lipid Bilayer Environment. For crystal-lization, membrane proteins are normally extracted from theirnative lipid membrane environment, and instead detergents ordetergent-lipid mixtures are used to solubilize and reconstitute



E


membrane proteins in vitro.3 As a starting point for the presentwork, we allowed the timolol bound β2AR-T4L complex (asgiven by the X-ray structure) to settle down in a lipid bilayerenvironment. Figure 2A represents the time evolution of theRMSD of the protein backbone. A major contribution toRMSD was observed to originate from the 159-residue longT4L (Figure 2A), which was completely exposed to water. Thetransmembrane part of the protein was very stable with RMSDless than 1.5 Å and reached a plateau within the first 50 ns ofsimulation. The protein also retained its overall secondarystructure content in a membrane during the entire simulationtime span (Figure S3).We also calculated the average backbone RMSD (inset in

Figure 2A) and average helical content (Figure 2B, shown incomparison to the crystal structure) of TM1−7 and the C-terminal short helix H8, separately. These results, along withthe superimposed snapshots taken from the last 50 ns of thetrajectory (Figure 2C) revealed that from the studiedtransmembrane segments, TM1 showed the largest deviation(1.26 ± 0.01 Å), especially at its N-terminal end. Though thisvalue is quite small. TM2 (RMSD being 0.63 ± 0.01 Å) and thelongest helix TM3 (RMSD 0.60 ± 0.01 Å) were observed to bepretty stable. The mobility of the lysozyme bound to theintracellular ends of TM5 and TM6 had a major contribution tothe calculated RMSD of these two helices (0.91 ± 0.01 Å and0.90 ± 0.01 Å, respectively). The flexible C-terminal helix-8(RMSD having a value of 0.43 ± 0.01 Å) contributed to thedeviation of TM7 (RMSD 0.77 ± 0.01 Å) mainly near thecytoplasmic end. Altogether, these results showed that theprotein relaxed in the membrane environment without anysignificant deviation from the crystal structure.

Step 2: Removal of Timolol Resulted in Fluctuationsand a Relative Opening of the Ligand-Binding Cavity,Suggesting That in the Apo-Receptor the BindingCavity Can Oscillate between Several Conformations.In the crystal structure, the partially inverse agonist timolol wasbound at the main ligand-binding cavity of the receptor locatedbetween the extracellular segments of the transmembranehelices 3, 5, 6, and 7 (Figure 3A). With the removal of timolol,the most prominent observation was the penetration of a largenumber of water molecules into the binding pocket (Figure3B). Water molecules migrated rapidly into the cavity in orderto fill the void generated by the removal of timolol. However,also in the presence of timolol the cavity was highly hydrated(Figure 3A), and we found on average ∼18 water molecules tobe present (at any moment) inside the pocket during the last100 ns of the trajectory. This level of hydration is in agreementwith a previous study, showing the fundamental role of water asa determinant of binding affinity of β-blockers to β2AR.44 Withthe removal of timolol, the number of water moleculesoccupying the cavity increased to ∼28 (these values are basedon calculations where we integrated the area under the numberdensity plot of water molecules that penetrated the bindingsite). Together with water, a sodium ion was observed to enterdeep into the cavity (Figure 3B) and stay there for the rest ofthe simulation time. The ion binding to the cavity is quitecertainly driven by electrostatics, since the cavity has anegatively charged D113, which attracts Na+, neutralizing thatregion.Next, we made a closer inspection of what happened to the

residues at the ligand-binding site (Figure 3C,D). The directβ2AR-timolol hydrogen bonding interactions were found to

Figure 2. Equilibration of the ligand-bound β2AR-T4L complex in a membrane (Method S). A) Time evolution of the backbone root-mean-squaredeviation (RMSD) of the entire β2AR-T4L complex (blue), β2AR-T4L without T4L (cyan), and the transmembrane region of the receptor (green).The inset represents the time average (the last 100 ns) for the RMSD of each individual transmembrane helix TM1-7 and helix-8. The error in theaverage RMSD per helix was less than 0.01 Å. B) The average number of residues with helical content (sum of α-helix, Π-helix, and 310-helix)calculated from the simulation trajectory (blue bars, average over the last 100 ns) and compared to the crystal structure 3D4S (red bars). C)Superimposed snapshots from the last 50 ns of the trajectory (blue ribbons, taken every 5 ns), with respect to the crystal structure (red ribbon). Theerror in the number of residues with helical content was below 0.04.



F



take place for (i) D1133.32 and N3127.39 interacting with thepolar tail of timolol, (ii) N2936.55 interacting with the timololhead (through the ether moiety of the morpholino ring), and(iii) T1183.37 hydrogen bonding with the thaidiazole ring oftimolol (Figure 3C). D1133.32, N3127.39, and N2936.55 were themost contributing residues, in line with previous experimentaland mutagenesis studies.13,45−47 In addition to these, weobserved a very persistent water-bridging interaction betweenS2075.46 and N of the thaidiazole ring of timolol (Figure 3C).Meanwhile, β2AR residues, which frequently involved innonbonded interactions with timolol are V114 and V117 inTM3, F193 in ECL2, and F289 and F290 as well as W286 inTM6. Other residues contributed less to these interactions(they interacted with timolol (measured as a contact betweenany heavy atom of timolol and the residue within 0.45 nm) forless than 10% of simulation time). Both in the presence andabsence of timolol, the polar triads D1133.32-Y3167.43-N3127.39

and S2045.43-N2936.55-Y3087.35 remained stabilized through anetwork of intraprotein hydrogen bonds (Figure 3C,D). Theseintraprotein interactions helped in stabilizing the trans-membrane helix interfaces (e.g., interfaces of TM3 and TM7in the case of D1133.32-Y3167.43-N3127.39 and TM domains 5−6−7 in the case of S2045.43-N2936.55-Y3087.35). As mentionedabove, a sodium ion entered the pocket after the removal oftimolol and bound near the negatively charged D1133.32. The

ion further stabilized the D1133.32-Y3167.43-N3127.39 anchor site,which was previously found to interact with the oxypropanol-amine tail of timolol (Figure 3C,D).With the removal of timolol, the main change we observed at

the ligand-binding cavity took place around S2075.46 in TM5.Serine residues S2035.42, S2045.43, and S2075.46 in TM5 areknown to play a key role in agonist binding. Previous studiesshowed that serine residues of TM5 form strong polarinteractions specifically with catecholamine agonists, whichresulted in an inward shift of the extracellular part of TM5around S2075.46, as compared to the antagonist-bound inactiveform.4,48,49 Here, to measure the displacement of TM5, wecalculated the distance between Cα atoms of S2075.46 andD1133.32 (as the position of D1133.32 in TM3 did not changenoticeably during the simulation). As shown in Figure 4, in thepresence of timolol, the ligand-binding site remained verystable, and there was only a small fluctuation in the S2075.46-D1133.32 distance around an average value of 12 Å. This value isin quite excellent agreement with the corresponding value(12.07 Å) obtained from the crystal structure of timolol-boundβ2AR.

13 However, as we removed the ligand, the pocket startedto oscillate between the closed and open forms with theS2075.46-D1133.32 distance ranging between 11 and 14.2 Å(Figure 4). Near the end of the simulation, the extracellular partof TM5 drifted away from D1133.32 in TM3, resulting in an

Figure 3. Interactions at the ligand-binding site (Method S). A) A representative simulation snapshot shows timolol at the main ligand-bindingcavity of β2AR. Timolol is shown as spheres and is colored according to the atom types (magenta for C, blue for N, red for O, yellow for S, and whitefor H). Nonpolar hydrogens are not displayed for clarity’s sake. Protein is represented by a green cartoon with transmembrane helix numbering.Waters inside the cavity are shown as a transparent cyan surface. B) A representative snapshot of the simulation after removal of timolol. Morewaters occupy the cavity along with a sodium ion (shown as a blue sphere). C−D) Close-up of polar interactions at the ligand-binding site:interactions between timolol and protein are shown by red dotted lines and intraprotein interactions by back-dotted lines. Amino acids participatingin the interaction are highlighted as sticks. Waters are not shown for clarity, except one (cyan bead), which made a water-bridge between Ser2075.46

and the thiadiazole ring N of timolol (panel C).



G



opening of the ligand binding site around S2075.46 by ∼2 Å.This outward movement of TM5 around S2075.46 took placeexactly in the opposite direction compared to the case ofagonist binding.Given the above, our results suggest that in the apo-receptor

the binding cavity can oscillate between various conformations.The presence of a ligand considerably stabilizes the pocket withfavorable receptor−ligand interactions and shifts the equili-brium toward a specific conformation. We will discuss this indetail in future publications.Step 3: With the Replacement of T4L by ICL3, the

Cytoplasmic End of TM6 Moved toward the Core of theReceptor, Further Closing the G-Protein Binding Site. Incrystallization, the incorporation of T4L stabilizes the dynamic

region of β2AR, namely the cytoplasmic ends of TM5 andTM6, and the intervening ICL3.4 The engineered β2AR-T4Lretains nearly all functions of the native receptor.50 Despitethat, ICL3 is known to directly interact with the G-protein andthus possibly affect receptor activation.51 In our study, the thirdICL was observed to be the most flexible part of the receptor,exhibiting the highest root-mean-square fluctuation (Figure5A). As Figure 5B depicts, ICL3 adopted multiple con-formations and orientations during the course of thesimulation, indicating its highly dynamic nature: starting froma relatively open initial structure (inset I at t = 0 ns in Figure5B), the loop first packed itself more tightly under theextracellular ends of the helix-bundle, near the G-protein’sbinding site (inset II at t = 60 ns), but later it again opened upand moved further away from the core of the receptor (inset IIIat t = 200 ns). In a recent study,11 a similar structure as in theinset II was pointed to be a “very inactive” conformation,backed up by a view that ICL3 almost completely blocks the G-protein’s binding site. During our simulation, the loopremained mostly unstructured. The formation of a shortsegment of antiparallel β-strands between the residues Q231-K232 and S261-S262 was most frequent with regard to thepossible secondary structure elements, complemented byoccasional formation of a short helix between the residues247−250 (Figure 5B). Additional simulations also supportedthe disordered and highly dynamic nature of ICL3 (see SI,Figures S4 and S5). These simulations also showed that theflexibility of ICL3 did not destabilize the rest of β2AR or affectits secondary structure.The mobility of ICL3 may affect the dynamics of the

transmembrane part, particularly helices 5 and 6. In the presentwork, after replacement of T4L by ICL3, we observed a shift of∼4 Å (compared to the crystal structure) of the intracellularend of TM6 toward the core of the heptahelical bundle,

Figure 4. Fluctuation and opening at the ligand-binding site afterremoval of timolol (Method S). Time evolution for the distancebetween Cα atoms of D1333.32 in TM3 and S2075.46 in TM5 of β2AR-T4L, with (light red) and without (light blue) timolol. Corresponding50-point running averages are shown in dark color. Black dashed linerepresents the S2075.46-D1133.32 distance in the crystal structure (PDBid: 3D4S).

Figure 5. Dynamics of ICL3 (Method S). A) Root-mean-square fluctuation (RMSF) of the backbone atoms per amino acid residues of β2AR-ICL3.B) Snapshots showing the various conformations and orientations adopted by ICL3 during the course of the simulation: inset I in the beginning ofthe simulation (0 ns), inset II at 60 ns, and inset III in the end of the simulation (200 ns). In the snapshots, the secondary structures are colored asmagenta for the α-helix, orange for the extended-β structure, cyan for turn, and white for coil.



H




followed by formation of the R1313.50-E2686.38 salt-bridge(Figure 6A,B). Such inward movement of TM6 largely closedthe G-protein binding site and occurred exactly in the oppositedirection than the displacement observed during receptoractivation. Figure 6C represents the displacement of thecytoplasmic end of TM6 from TM3, and the time evolution ofsalt-bridge formation during the simulation time span. Ourresults show that in β2AR-ICL3, the intracellular end of TM6approached TM3, without noticeable displacement of the latter.As a result, the corresponding Cα-Cα distance of R1313.50-E2686.38 decreased by ∼2 Å compared to the crystal structure(Figure 6). Unlike β2AR with the lysozyme (β2AR-T4L), thepresence of a flexible ICL3 (replacing T4L) allowed the twohelices to be in close proximity and also fostered the formationof a salt-bridge between them. For β2AR-ICL3, we observedthat as the R1313.50-E2686.38 salt-bridge formed (in ∼80 ns), itremained very stable until the end of the simulation (Figure6C). The salt-bridge formation is in agreement with a previoussimulation study10,52 and based on the available mutagenesisdata53 is known to play a crucial role in β2AR activation.In relation to the conformational change in the intracellular

side of the receptor, we also monitored changes that occurredin the extracellular ligand-binding site (using the same approachas described in Step 2). As shown in Figure S6, in β2AR-ICL3the S2075.46-D1133.32 distance was ∼14 Å, which is ∼2 Å largerthan the crystal structure value. As compared to β2AR-T4L, inβ2AR-ICL3 there was no further opening at the ligand-bindingpocket or oscillation between the closed and open forms(Figures 4 and S6). Thus, the narrower G-protein binding site

in β2AR-ICL3 seems to stabilize the ligand-binding site to arelatively open state, which supports the view that the allostericnetwork acts between the two distal sites of β2AR.

Step 4: Back Mutation of W1223.41 to E1223.41 Affectsthe Packing and Interactions at the TransmembraneHelix Interface. In β2AR, E122

3.41 of TM3 is located at theinterface between the helices TM3-TM5, facing the membraneexterior. Engineering the helical interface by mutating E1223.41

with bulky hydrophobic residues, e.g., tryptophan (as found inrhodopsin) has shown a significant potential to enhancereceptor stability while maintaining wild-type functions.54 Asfound in our simulation, W1223.41 was sandwiched betweenV160 in TM4 and P211 in TM5 (Figure 7A). The aromaticring of W1223.41 formed CH-π stacking interactions with thering of P211. The indole nitrogen of W1223.41 formed a directhydrogen bond with the hydroxyl group of T164 in TM4(Figure 7A), which further stabilized helix 4. However, we didnot observe any polar interactions between W1223.41 N andbackbone carbonyl of Val206 in TM5, as previously predictedbased on homology modeling.54

In our studies, back mutation of W1223.41 to E1223.41

disrupted the above-mentioned packing interactions betweenthe TM segments (Figure 7A). Interaction with P211 was lost.The carboxyl group of E1223.41 resides too far to make directhydrogen bonding with T164. Rather, E1223.41 formed onlyweak water-mediated interactions with T164 in TM4 and withthe backbone carbonyl group of V206 in TM5 (Figure 7A).

Figure 6. Changes at the G-protein binding site after removal of T4L (Method S). The snapshots of β2AR-ICL3 (magenta) represent thedisplacement of the cytoplasmic end of TM6 toward TM3, compared to the crystal structure (orange). A) View from the G-protein binding site andB) side view. The cytoplasmic half of TM3 and TM6 is highlighted in the snapshot. The residues R1313.50 and E2686.38 of the simulation model areshown as sticks and are colored according to atom types (green for C, red for O, blue for N). Hydrogen atoms are not displayed for clarity. Thecorresponding residues of the crystal structure are shown as orange sticks and exhibit a broken salt-bridge. In the snapshot A) the gray arrow showsthe shift of the intracellular end of TM6 (Cα atom of K267 shown as a sphere) of β2AR-ICL3 (shown in magenta) compared to the crystal structure(shown in orange). C) Upper panel shows the distances between the Cα atoms R1313.50 and E2686.38 in β2AR-T4L (gray) and β2AR-ICL3(magenta). The lower panel shows the minimum distance between the guanidium nitrogen atoms of R1313.50 and the carboxyl oxygen atoms ofE2686.38, characterizing salt bridge formation. Dark colors refer to corresponding 50-point running averages. The orange lines refer to thecorresponding values in the crystal structure [PDB id: 3D4S].



I



Altogether the results point to a conclusion that the loss ofpacking interactions at the transmembrane helix interfacecontributes to the inherent structural instability of β2AR.Step 5: Back Mutation of E1875.26 to N1875.26 Did Not

Affect Considerably the Local Structure and Orientationof ECL2. Human β2AR has a potential N-glycosylation sitelocated in the second extracellular loop (ECL2) at the residue1875.26.55 For structure determination in experiments, N187was mutated to E187 to avoid glycosylation and therefore fostercrystal formation. We examined whether the mutation ofN187E to E187 had an impact on the structure/orientation ofECL2.As shown in Figure 7B, the local structure remained largely

the same after the reverse mutation of E1875.26 back to thenative N1875.26. The short helical region of ECL2 remainedintact. T1895.28 on ECL2, which was previously interacting withthe carboxyl oxygen of E1875.26, was now found to interact withthe amide oxygen of N1875.26. For both cases, the intraloopdisulfide bond between C1844.76 and C1905.29 stabilized theconformation of ECL2. Another disulfide bond betweenC1915.30 and C1063.25 in TM3 maintained the orientation ofECL2 with respect to the transmembrane helix-bundle (Figure7B). Therefore, the mutation alone had no significant effect onECL2, which acts as an access point to the binding site.In human β2AR the N-glycosylated residues are known to

play an important role for the proper insertion of the receptorin a cell membrane and also in agonist trafficking.55,56 However,understanding the effect of glycosylation is beyond the scope ofthe current study and remains to be discussed elsewhere.Step 6: The Stepwise Approach Resulted in an

Improved Model, Which Reproduced the Properties ofthe Native Receptor Very Well. Figure 8 shows the keyproperties of the receptor structure found through Method S, in

which the modifications done for experimental structuredetermination were removed step by step. These findingsshould be compared with the results presented in Figure 1 forMethod A, where the same structural modifications werecorrected simultaneously in a single step. Below we discuss theresults of Method S, unless said otherwise.When we summarize the outcome of Method S, we first find

(Figure 8A) that the backbone RMSD profile of the TM regionof β2AR reached a plateau much faster (within the first 50 ns ofsimulation) than the entire receptor (∼170 ns). The averageRMSD (the last 100 ns) of the TM region (backbone atomsonly) from the crystal structure was 1.45 ± 0.01 Å, while thecorresponding RMSD in Method A was 2.89 ± 0.01 Å.Figure 8B shows the average RMSDs per TM residues. In

Method S, the majority of the protein exhibited a deviation thatwas less than 1.5 Å from the crystal structure, considerably lessthan what was found in Method A (Figure 8B). With Method Sthe N-terminal end of TM1 was the major fluctuating part ofthe protein (RMSD < 3 Å), although the drift was muchsmaller than in the model based on Method A (>6 Å) (Figure8B). TM2 and the longest TM3 were the most stable onesamong all the helices. TM4 fluctuated in both models (thoughthe results for RMSD per residue were smaller in the case ofMethod S). The smallest helix TM4 is considered to be theweakest point (in terms of exhibiting the lowest intramolecularpacking calculated by the occluded surface area method in ref13) of the β2AR fold.13 Unlike in Method A, in Method S wedid not observe any significant deviation at the intracellular endof TM3. Instead the intracellular end of TM6 (residues 267−269) exhibited larger deviations (>3 Å). This occurred due to ashift of TM6 toward the core of the receptor (after removal ofT4-lysozyme), in order to form a salt bridge with thecytoplasmic end of TM3 (Figure 6A). Similarly, the fluctuations

Figure 7. Effects of mutations (Method S). A) Representative simulation snapshots showing the interactions at the TM4-TM3-TM5 interfacearound W1223.41/E1223.41 of TM3. In the figure W1223.41/E1223.41 residues are represented as spheres and are colored according to the atom type(cyan for C, blue for N, red for O, yellow for S, and white for H). The neighboring residues are shown as sticks. A water molecule within 5 Å aroundthe 3.41 position is shown as a ball-and-stick representation. Hydrogen bonding interactions are shown with purple dashed lines. B) Representativesimulation snapshots showing the interactions around the residue 187 at ECL2 (represented as spheres and colored according to the atom type).Two disulfide bonds of ECL2, C1844.76-C1905.29 and C1063.25-C1915.30, are shown as sticks.



J



around S2075.46 in TM5 (Figure 4, after removing the timololligand) were reflected in higher RMSD values of this region(Figure 8B). The deviation observed in the cytoplasmic end ofTM7 (at the conserved NPxxY motif) was consistent with anearlier study.7

As shown in Figures 8C and S7, the β2AR model obtained byMethod S retained its overall secondary structure whensimulated in a DOPC lipid bilayer. All TM domains exhibitedvery high helical propensity: helical content of TM1, TM2,TM3, TM5, and TM7 were above 80%, while that of TM4 andTM6 were ∼75% (Figure 8D and Table S2). Proline residuescaused local opening of a helix at TM4 (P168) and TM7(P323). As compared to Method A (Long simulation), themost noticeable enhancement of helical content was observedfor TM1 (from 52% in Method A to 84% in Method S), TM2(from 65% in Method A to 86% in Method S), TM5 (from60% in Method A to 82% in Method S), TM6 (from 59% inMethod A to 75% in Method S), and TM7 (from 71% inMethod A to 84% in Method S) (Figure 8D and Table S2).Similar conclusions were drawn from the results by the differentsequence of steps in Control-2 (Figure 8D, Figure S8, andTable S2), which indicated that the order of steps did not affectthe results of Method S to a significant extent.

The β2AR model obtained by Method S was further validatedin a DOPC bilayer with 10 mol % cholesterol (Control-1).Cholesterol is the most prevalent sterol of the plasmamembrane and known to influence GPCRs structure/function.21,22 As shown in Figures 8B and S9, per residueRMSDs of TM residues showed similar behavior in bothbilayers. The high helical contents of TM domains as observedin a DOPC bilayer without cholesterol were well reproduced ina DOPC bilayer with 10 mol % cholesterol (Figure 8D andTable S2). To assess the effect of cholesterol on the stabilityand function of β2AR, more thorough studies are needed.Authors are planning to address this issue in the future.Summarizing, the present model using Method S showed

significant improvement in the total average helical content ofTM domains of β2AR compared to the model prepared withMethod A (Figure 8D). For Method S, the helical structure ofTM region of the receptor was well preserved both in DOPCbilayer and in more biologically relevant cholesterol-containingbilayer. The stepwise method was able to reproduce theconserved heptahelical architecture of GPCR and provided amuch better receptor model than Method A.

Figure 8. Properties of the native receptor model generated through the stepwise approach (Method S). A) Root-mean-square deviation (RMSD) ofthe protein backbone: entire β2AR (black) and the transmembrane region (cyan), throughout the simulation. B) Per residue RMSD (average overthe last 100 ns) of the simulated receptor model compared to the crystal structure, calculated over the backbone atoms of the transmembrane region.In the graph, the cyan and gray areas correspond to the stepwise approach (Method S) and the simultaneous removal of crystallization modifications(Method A), respectively. The snapshots show the front (TM1-TM4) and back (TM5-TM7) views of the protein’s average structure (over the last100 ns), with TM residues colored according to their respective RMSD values, and the rest of the protein represented as a transparent gray cartoon.C) A representative snapshot of the receptor with TM residues colored according to the secondary structure content (α-helix in purple, 310-helix inblue, turn in green, coil in orange): front (TM1-TM4) and back (TM5-TM7) sides. D) The number of residues with helical content (sum of α-, 310-,and Π-helices) for each TM helix calculated from different Method S trajectories (averaged over the last 100 ns): in a one-component DOPC bilayer(green), in a DOPC bilayer with 10 mol % cholesterol (cyan, Control-1), and in a DOPC bilayer with the reverted order of steps (yellow; Control-2). The results of Method S were compared to that found in Method A (red, Long simulation) and in the crystal structure 3D4S (gray). The error inthe number of residues with helical content was below 0.04.



K



■ CONCLUSIONS

In the present work, we used atomistic molecular dynamicssimulations to model the properties of the native β2AR receptorin a lipid membrane, starting from a heavily engineered crystalstructure. β2AR is one of the most well studied GPCRs, andthanks to recent advancements in development of crystal-lization techniques, several high-resolution structures of thisreceptor or GPCRs in general are now available in proteinstructure databases. Like many other GPCRs, the inactive β2AR(that is of our interest here) underwent several major structuralmodifications during its structure determination that weremade in experiments to foster crystallization, e.g., replacementof highly dynamic loops with a rigid lysozyme, introduction ofstabilizing mutations, etc. In order to overcome the inherentdynamic nature of the receptor, such structural modificationshave become general strategies for crystallizing membraneproteins.In the current study, we explored what could be the most

efficient way to get rid of the crystallization modificationswithout altering the properties of the native β2AR receptorsubstantially. By following the most usual approach, we firstremoved all changes made originally for crystallization and thensimulated the receptor in a lipid membrane environment. Inthis approach (Method A), we simultaneously modified theligand-binding site by removing timolol, altered the G-proteinbinding site by replacing T4L with ICL3, and also modified thetransmembrane helical interface by removing a mutation. Withall these simultaneous changes, the method led to a receptorstructure that deviated quite significantly from the crystalstructure. The major drawback found in this method was that itresulted in a considerable drop in the helical content of thetransmembrane region of β2AR, e.g., on average the helicalpercentage decreased (compared to helicity in the crystalstructure) to 58% in TM1, 56% in TM6, and 62% in TM7.There is reason to stress that heptahelical architecture is aninherent property of GPCRs, and it should be reproduced inthe computational model, too.In an alternative approach (Method S), we removed the

modifications (made for crystallization experiments) in a morecareful, stepwise manner (one at a time, each phase followed byhundreds of nanoseconds of equilibration). In this method, theinstability in the receptor structure was overcome, and thehelical characteristic of the transmembrane region waspreserved. Without doubt, the results of Method S are quitesuperior compared to those given by the more straightforwardtechnique, Method A.The price one has to pay to use Method S is the increased

computational cost. In the present study, the simulations withMethod S covered a total simulation time scale of about 1.5 μs.Meanwhile, simulations with Method A could have been donein about 250 ns (though we extended those simulations to amicrosecond). Yet, we consider that the accuracy and thereliability of the simulation model and its predictions justify theadded computational burden. Further, while one may debatewhether the results of Methods S and A can be compared toeach other on equal footing due to the different simulationtimes, it is important to bring out that many membrane proteinsimulations are performed over times of the order of 100 ns.This is particularly common, e.g., with the CHARMM forcefield that is often considered the force field of choice for proteinsimulation. The simulation times used in this work (∼5 μs in

total) should therefore provide a reasonable benchmark for thesimulations in the field in general.Importantly, we did several control simulations to back up

the validity of this conclusion. We tested how the order ofdoing the structural modifications to get back to the pristineprotein affecting the results. Our simulations showed that theorder of steps does not affect the main results of Method S to asignificant degree. We also tested the influence of cholesteroland found that it does not play a significant role in thecomparison between Methods A and S. We also tested that thepreparation of the loop structures in the model systems did notaffect the results much. Finally, the restraints used for theprotein were in this work released in a standard way during atime scale of 50 ns. A more gradual release of restraints mightimprove the results slightly.Our results also provided a considerable amount of

knowledge regarding the effects of individual structuralmodifications, such as mutations and the use of antibodies inprotein structure. The rigid T4L structure was fused betweenTM5-TM6 in experiment to replace the inherently flexibleICL3. T4-lysozyme also restricted the dynamics of TM5 andTM6, which is known to be important for receptor activationand responsible for the open structure of G-protein binding siteobserved in the crystal structure of β2AR. The replacement ofT4L with ICL3 relatively closed the G-protein binding site, asthe intracellular end of TM6 moved 4 Å toward the core of thehelical bundle. Meanwhile, removal of timolol resulted inrelative opening of the ligand-binding cavity. The changesobserved in the distal ligand- and G-protein binding sitesaccounted for the allosteric regulation in GPCRs.The reformation of the R1313.50-E2686.38 salt bridge, which

was initially broken in the crystal structure of inactive β2AR, hasbeen reported in previous computational studies.10,52 Dror et al.described the carazolol-bound β2AR conformation with anintact R1313.50-E2686.38 salt bridge as an alternative inactiveconformation of β2AR, distinct from the experimentallyreported structure.10 This study, however, did not focus onthe change at the ligand-binding site.10 For comparison, in mostof the previous studies, the receptor models used in simulationswere modified to some extent. In many of these studies, afterremoving T4L, instead of incorporating ICL3 the newlyexposed ends at the cytoplasmic side of TM5 and TM6 wereeither left open7,57,58 or clipped to each other via a peptidebond10,52 (termed a “clipped model”). A recent work by Ozcanet al. showed that ICL3 plays a significant role in the intrinsicconformational dynamics of β2AR and that the clipped receptormodel exhibited very different dynamical behavior than themodel with ICL3.11 Additionally, in most of these simulations,the N1875.26E mutation (as found in the crystal structure) waskept unaltered.7,10,11,52,58 The purpose of the present work wasto model the receptor as closely as possible to its biologicallyrelevant form and also to characterize how much themodifications made in crystallization can actually affect theintrinsic structural properties of the native receptor.Concluding, our results emphasize how sensitively mem-

brane protein structure and dynamics depend on thepreparation of the protein whose properties one would liketo explore through atomistic simulations. If the preparation isdone in a straightforward manner by cutting corners, then theoutcome of the simulations can be partly artificial, notcorresponding to the true native state under equilibriumconditions. This threat is quite possible especially in shortsimulations where the protein has not enough time to relax.



L


Given that many of the receptor structural changes take placeover long time scales (ranging from microseconds to muchlarger times), and considering that this is particularly the casefor receptors such as GPCRs where structural changes typicallytake place through the entire protein, there is reason to warrantgreat care in starting atomistic simulations of membraneproteins whose crystal structure determination has requiredconsiderable structural modifications. Altogether, we considerthat the protocol described here (Method S) may offer a usefulstrategy for simulating a variety of native state receptors, whosecrystal structures suffer from similar modifications.

■ ASSOCIATED CONTENT*S Supporting InformationFurther results for the protein structures. The SupportingInformation is available free of charge on the ACS Publicationswebsite at DOI: 10.1021/acs.jctc.5b00070.

■ AUTHOR INFORMATIONCorresponding Author*E-mail: [email protected] authors declare no competing financial interest.

■ ACKNOWLEDGMENTSReinis Danne is thanked for technical support at the early stageof this work. CSC − Finnish IT Centre for ScientificComputing (Espoo, Finland) is acknowledged for computerresources granted through the Grand Challenge projectconcept. European Research Council (Advanced Grant projectCROWDED-PRO-LIPIDS) and the Academy of FinlandCenter of Excellence program are thanked for financial support.

■ REFERENCES(1) Lefkowitz, R. J. Seven transmembrane receptors - A brief personalretrospective. Biochim. Biophys. Acta, Biomembr. 2007, 1768, 748−755.(2) Lefkowitz, R. J. The superfamily of heptahelical receptors. Nat.Cell Biol. 2000, 2, E133−E136.(3) Seddon, A. M.; Curnow, P.; Booth, P. J. Membrane proteins,lipids and detergents: Not just a soap opera. Biochim. Biophys. Acta,Biomembr. 2004, 1666, 105−117.(4) Kobilka, B. The structural basis of G-protein-coupled receptorsignaling (Nobel Lecture). Angew. Chem., Int. Ed. 2013, 52, 6380−6388.(5) Tate, C. G. A crystal clear solution for determining G-protein-coupled receptor structures. Trends Biochem. Sci. 2012, 37, 343−352.(6) Poyry, S.; Cramariuc, O.; Postila, P. A.; Kaszuba, K.; Sarewicz, M.;Osyczka, A.; Vattulainen, I.; Rog, T. Atomistic simulations indicatecardiolipin to have an integral role in the structure of the cytochromebc1 complex. Biochim. Biophys. Acta, Bioenergetics 2013, 1827, 769−778.(7) Dror, R. O.; Arlow, D. H.; Maragakis, P.; Mildorf, T. J.; Pan, A.C.; Xu, H.; Borhani, D. W.; Shaw, D. E. Activation mechanism of theβ2-adrenergic receptor. Proc. Natl. Acad. Sci. U. S. A. 2011, 108,18684−18689.(8) Li, Y. Y.; Hou, T. J.; Goddard, W. A., III Computational modelingof structure-function of G protein-coupled receptors with applicationsfor drug design. Curr. Med. Chem. 2010, 17, 1167−1180.(9) G-Protein Coupled Receptors − Modeling and Simulation. Series:Advances in Experimental Medicine and Biology; Filizola, M. G., Ed.;Springer: 2014; Vol. 796 .(10) Dror, R. O.; Arlow, D. H.; Borhani, D. W.; Jensen, M. Ø.; Piana,S.; Shaw, D. E. Identification of two distinct inactive conformations ofthe β2-adrenergic receptor reconciles structural and biochemicalobservations. Proc. Natl. Acad. Sci. U. S. A. 2009, 106, 4689−4694.

(11) Ozcan, O.; Uyar, A.; Doruker, P.; Akten, E. D. Effect ofintracellular loop 3 on intrinsic dynamics of human β2-adrenergicreceptor. BMC Struct. Biol. 2013, 13, 29.(12) Grossfield, A. Recent progress in the study of G protein-coupledreceptors with molecular dynamics computer simulations. Biochim.Biophys. Acta, Biomembr. 2011, 1808, 1868−1878.(13) Hanson, M. A.; Cherezov, V.; Griffith, M. T.; Roth, C. B.;Jaakola, V.-P.; Chien, E. Y. T.; Velasquez, J.; Kuhn, P.; Stevens, R. C. Aspecific cholesterol binding site is established by the 2.8 Å structure ofthe human β2-adrenergic receptor. Structure 2008, 16, 897−905.(14) Rasmussen, S. G. F.; Choi, H.-J.; Rosenbaum, D. M.; Kobilka, T.S.; Thian, F. S.; Edwards, P. C.; Burghammer, M.; Ratnala, V. R. P.;Sanishvili, R.; Fischetti, R. F.; Schertler, G. F. X.; Weis, W. I.; Kobilka,B. K. Crystal structure of the human β2 adrenergic G-protein-coupledreceptor. Nature 2007, 450, 383−U384.(15) Cherezov, V.; Rosenbaum, D. M.; Hanson, M. A.; Rasmussen, S.G. F.; Thian, F. S.; Kobilka, T. S.; Choi, H.-J.; Kuhn, P.; Weis, W. I.;Kobilka, B. K.; Stevens, R. C. High-resolution crystal structure of anengineered human β2-adrenergic G protein-coupled receptor. Science2007, 318, 1258−1265.(16) Rasmussen, S. G. F.; Choi, H.-J.; Fung, J. J.; Pardon, E.;Casarosa, P.; Chae, P. S.; DeVree, B. T.; Rosenbaum, D. M.; Thian, F.S.; Kobilka, T. S.; Schnapp, A.; Konetzki, I.; Sunahara, R. K.; Gellman,S. H.; Pautsch, A.; Steyaert, J.; Weis, W. I.; Kobilka, B. K. Structure of ananobody-stabilized active state of the β(2) adrenoceptor. Nature2011, 469, 175−180.(17) Rasmussen, S. G. F.; DeVree, B. T.; Zou, Y.; Kruse, A. C.;Chung, K. Y.; Kobilka, T. S.; Thian, F. S.; Chae, P. S.; Pardon, E.;Calinski, D.; Mathiesen, J. M.; Shah, S. T. A.; Lyons, J. A.; Caffrey, M.;Gellman, S. H.; Steyaert, J.; Skiniotis, G.; Weis, W. I.; Sunahara, R. K.;Kobilka, B. K. Crystal structure of the β2 adrenergic receptor-Gsprotein complex. Nature 2011, 477, 549−U311.(18) Rog, T.; Vattulainen, I. Cholesterol, sphingolipids, andglycolipids: What do we know about their role in raft-like membranes?Chem. Phys. Lipids 2014, 184, 82−104.(19) Neuvonen, M.; Manna, M.; Mokkila, S.; Javanainen, M.; Rog,T.; Liu, Z.; Bittman, R.; Vattulainen, I.; Ikonen, E. Enzymatic oxidationof cholesterol: Properties and functional effects of cholestenone in cellmembranes. PLoS One 2014, 9, e103743.(20) Manna, M.; Mukhopadhyay, C. Cholesterol driven alteration ofthe conformation and dynamics of phospholamban in modelmembranes. Phys. Chem. Chem. Phys. 2011, 13, 20188−20198.(21) Paila, Y. D.; Chattopadhyay, A. The function of G-proteincoupled receptors and membrane cholesterol: Specific or generalinteraction? Glycoconjugate J. 2009, 26, 711−720.(22) Zocher, M.; Zhang, C.; Rasmussen, S. G. F.; Kobilka, B. K.;Muller, D. J. Cholesterol increases kinetic, energetic, and mechanicalstability of the human β2-adrenergic receptor. Proc. Natl. Acad. Sci. U.S. A. 2012, 109, 3463−4372.(23) Kulig, W.; Tynkkynen, J.; Javanainen, M.; Manna, M.; Rog, T.;Vattulainen, I.; Jungwirth, P. How well does cholesteryl hemisuccinatemimic cholesterol in saturated phospholipid bilayers? J. Mol. Model.2014, 20, 2121.(24) Javanainen, M. Universal method for embedding proteins intocomplex lipid bilayers for molecular dynamics simulations. J. Chem.Theory Comput. 2014, 10, 2577−2582.(25) Ballesteros, J. A.; Weinstein, H. Integrated methods for theconstruction of three dimensional models and computational probingof structure function relations in G protein-coupled receptors. MethodsNeurosci. 1995, 25, 366−428.(26) The UniProt Consortium, Activities at the universal proteinresource (UniProt). Nucleic Acids Res. 2014, 42, D191−198.(27) Sali, A.; Blundell, T. L. Comparative protein modeling bysatisfaction of spatial restraints. J. Mol. Biol. 1993, 234, 779−815.(28) Jaakola, V. P.; Prilusky, J.; Sussman, J. L.; Goldman, A. Gprotein-coupled receptors show unusual patterns of intrinsic unfolding.Protein Eng., Des. Sel. 2005, 18, 103−110.(29) Jorgensen, W. L.; Maxwell, D. S.; TiradoRives, J. Developmentand testing of the OPLS all-atom force field on conformational



M

http://pubs.acs.org

http://pubs.acs.org

http://pubs.acs.org/doi/abs/10.1021/acs.jctc.5b00070

mailto:[email protected]


energetics and properties of organic liquids. J. Am. Chem. Soc. 1996,118, 11225−11236.(30) Kaminski, G. A.; Friesner, R. A.; Tirado-Rives, J.; Jorgensen, W.L. Evaluation and reparametrization of the OPLS-AA force field forproteins via comparison with accurate quantum chemical calculationson peptides. J. Phys. Chem. B 2001, 105, 6474−6487.(31) Maciejewski, A.; Pasenkiewicz-Gierula, M.; Cramariuc, O.;Vattulainen, I.; Rog, T. Refined OPLS all-atom force field for saturatedphosphatidylcholine bilayers at full hydration. J. Phys. Chem. B 2014,118, 4571−4581.(32) Jorgensen, W. L.; Chandrasekhar, J.; Madura, J. D.; Impey, R.W.; Klein, M. L. Comparison of simple potential functions forsimulating liquid water. J. Chem. Phys. 1983, 79, 926−935.(33) Bayly, C. I.; Cieplak, P.; Cornell, W. D.; Kollman, P. A. A well-behaved electrostatic potential based method using charge restraintsfor deriving atomic charges - the RESP model. J. Phys. Chem. 1993, 97,10269−10280.(34) Hess, B.; Kutzner, C.; van der Spoel, D.; Lindahl, E. GROMACS4: Algorithms for highly efficient, load-balanced, and scalablemolecular simulation. J. Chem. Theory Comput. 2008, 4, 435−447.(35) Bussi, G.; Donadio, D.; Parrinello, M. Canonical samplingthrough velocity rescaling. J. Chem. Phys. 2007, 126.(36) Parrinello, M.; Rahman, A. Polymorphic transitions in single-crystals - A new molecular-dynamics method. J. Appl. Phys. 1981, 52,7182−7190.(37) Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. LINCS: Alinear constraint solver for molecular simulations. J. Comput. Chem.1997, 18, 1463−1472.(38) Essmann, U.; Perera, L.; Berkowitz, M. L.; Darden, T.; Lee, H.;Pedersen, L. G. A smooth particle mesh Ewald method. J. Chem. Phys.1995, 103, 8577−8593.(39) Lehtonen, J. V.; Still, D. J.; Rantanen, V. V.; Ekholm, J.;Bjorkland, D.; Iftikhar, Z.; Huhtala, M.; Repo, S.; Jussila, A.; Jaakkola,J.; Pentikainen, O.; Nyronen, T.; Salminen, T.; Gyllenberg, M.;Johnson, M. S. BODIL: A molecular modeling environment forstructure-function analysis and drug design. J. Comput.-Aided Mol. Des.2004, 18, 401−419.(40) Kabsch, W.; Sander, C. Dictionary of protein secondarystructure: Pattern recognition of hydrogen-bonded and geometricalfeatures. Biopolymers 1983, 22, 2577−2637.(41) Armen, R.; Alonso, D. O. V.; Daggett, V. The role of alpha-,3(10)-, and pi-helix in helix -> coil transitions. Protein Sci. 2003, 12,1145−1157.(42) Lazarova, T.; Brewin, K. A.; Stoeber, K.; Robinson, C. R.Characterization of peptides corresponding to the seven trans-membrane domains of human adenosine A2a receptor. Biochemistry2004, 43, 12945−12954.(43) Corin, K.; Baaske, P.; Ravel, D. B.; Song, J.; Brown, E.; Wang,X.; Geissler, S.; Wienken, C. J.; Jerabek-Willemsen, M.; Duhr, S.;Braun, D.; Zhang, S. A Robust and rapid method of producing soluble,stable, and functional G-protein coupled receptors. PLoS One 2011, 6,e23036.(44) Kaszuba, K.; Rog, T.; Bryl, K.; Vattulainen, I.; Karttunen, M.Molecular dynamics simulations reveal fundamental role of water asfactor determining affinity of binding of β-blocker nebivolol to β2-adrenergic receptor. J. Phys. Chem. B 2010, 114, 8374−8386.(45) Strader, C. D.; Sigal, I. S.; Candelore, M. R.; Rands, E.; Hill, W.S.; Dixon, R. A. F. Conserved aspartic acid residues 79 and 113 of theβ-adrenergic-receptor have different roles in receptor function. J. Biol.Chem. 1988, 263, 10267−10271.(46) Suryanarayana, S.; Kobilka, B. K. Amino acid substitutions atposition 312 in the 7th hydrophobic segment of the beta2-adrenergicreceptor modify ligand-binding specificity. Mol. Pharmacol. 1993, 44,111−114.(47) Wieland, K.; Zuurmond, H. M.; Krasel, C.; Ijzerman, A. P.;Lohse, M. J. Involvement of Asn-293 in stereospecific agonistrecognition and in activation of the β2-adrenergic receptor. Proc.Natl. Acad. Sci. U .S. A. 1996, 93, 9276−9281.

(48) Strader, C. D.; Candelore, M. R.; Hill, W. S.; Sigal, I. S.; Dixon,R. A. F. Identification of 2 serine residues involved in agonist activationof the β-adrenergic-receptor. J. Biol. Chem. 1989, 264, 13572−13578.(49) Katritch, V.; Reynolds, K. A.; Cherezov, V.; Hanson, M. A.;Roth, C. B.; Yeager, M.; Abagyan, R. Analysis of full and partialagonists binding to β2-adrenergic receptor suggests a role oftransmembrane helix V in agonist-specific conformational changes. J.Mol. Recognit. 2009, 22, 307−318.(50) Rosenbaum, D. M.; Cherezov, V.; Hanson, M. A.; Rasmussen, S.G. F.; Thian, F. S.; Kobilka, T. S.; Choi, H.-J.; Yao, X.-J.; Weis, W. I.;Stevens, R. C.; Kobilka, B. K. GPCR engineering yields high-resolutionstructural insights into β2-adrenergic receptor function. Science 2007,318, 1266−1273.(51) Liggett, S. B.; Caron, M. G.; Lefkowitz, R. J.; Hnatowich, M.Coupling of a mutated form of the human β2-adrenergic receptor to Giand Gs. J. Biol. Chem. 1991, 266, 4816−4821.(52) Romo, T. D.; Grossfield, A.; Pitman, M. C. Concertedinterconversion between ionic lock substates of the β2 adrenergicreceptor revealed by microsecond timescale molecular dynamics.Biophys. J. 2010, 98, 76−84.(53) Ballesteros, J. A.; Jensen, A. D.; Liapakis, G.; Rasmussen, S. G.F.; Shi, L.; Gether, U.; Javitch, J. A. Activation of the β2-adrenergicreceptor involves disruption of an ionic lock between the cytoplasmicends of transmembrane segments 3 and 6. J. Biol. Chem. 2001, 276,29171−29177.(54) Roth, C. B.; Hanson, M. A.; Stevens, R. C. Stabilization of thehuman β2-adrenergic receptor 4−3-5 helix interface by mutagenesis ofGlu-1223.41, a critical residue in GPCR structure. J. Mol. Biol. 2008,376, 1305−1319.(55) Mialet-Perez, J.; Green, S. A.; Miller, W. E.; Liggett, S. B. Aprimate-dominant third glycosylation site of the β2-adrenergic receptorroutes receptors to degradation during agonist regulation. J. Biol.Chem. 2004, 279, 38603−38607.(56) Rands, E.; Candelore, M. R.; Cheung, A. H.; Hill, W. S.; Strader,C. D.; Dixon, R. A. F. Mutational analysis of β-adrenergic-receptorglycosylation. J. Biol. Chem. 1990, 265, 10759−10764.(57) Cang, X.; Du, Y.; Mao, Y.; Wang, Y.; Yang, H.; Jiang, H.Mapping the functional binding sites of cholesterol in β2-adrenergicreceptor by long-time molecular dynamics simulations. J. Phys. Chem.B 2013, 117, 1085−1094.(58) Xiao, X.; Zeng, X.; Yuan, Y.; Gao, N.; Guo, Y.; Pu, X.; Li, M.Understanding the conformation transition in the activation pathwayof β2 adrenergic receptor via a targeted molecular dynamicssimulation. Phys. Chem. Chem. Phys. 2014, 17, 2512−2522.



N


Documents

How to minimize artifacts in atomistic simulations of membrane proteins, whose crystal structure is heavily engineered: beta-2-adrenergic receptor in the spotlight